一句话：提出 META²V2V 元变形测试框架，用 V2V 信息扰动和多目标搜索检测多车 ADS 的健全性与鲁棒性缺陷。
在多车自动驾驶场景中，配备 ADS 的车辆通过 V2V 通信共享实时信息并基于相互感知决策，但共享信息受扰动时可能导致行为偏差。本文提出元变形测试框架 META²V2V，向 V2V 数据流系统注入四类受控常见扰动，观察其对车辆相互感知与后续交互的影响。框架从健全性与鲁棒性两方面定义 S-MR 和 R-MR：前者要求完美信息下性能应优于扰动下，后者要求信息扰动下性能不应过度退化。为高效生成违反上述关系的测试组，META²V2V 采用双方向多目标搜索测试。实验表明，相比随机测试，其检测到的健全性与鲁棒性违反分别多 733.3% 和 133.0%。
In next-generation traffic, multiple intelligent vehicles equipped with autonomous driving systems (ADS) operate simultaneously. These vehicles share real-time information through vehicle-to-vehicle (V2V) communication during operation and make decisions based on this mutual perception. In this paper, we present a novel METAmorphic testing framework to evaluate ADS performance in multi-vehicle scenarios, addressing the META information shared during V2V communication (META2V2V). META2V2V interacts with the shared information by systematically injecting four types of controlled common perturbations into V2V data streams and observing how these disruptions affect the vehicles’ mutual perception and subsequent interactions. META2V2V assesses ADS decisions from two aspects: soundness and robustness, by introducing soundness metamorphic testing relations (S-MR s) and robustness metamorphic relations (R-MR s), respectively. Specifically, S-MR s claim that the performance of ADS under perfect information should outperform that under perturbations, whereas R-MR s claim that the performance of ADS should not degrade too much under information perturbations. Furthermore, META2V2V employs multi-objective search-based testing conducted from two directions to efficiently generate test groups that violate S-MR s and R-MR s, respectively. Experimental results demonstrate that, compared to random testing, our approach can identify 733.3% and 133.0% more violations that reveal soundness and robustness issues.
@inproceedings{paperbot11767,
title = {META²V2V: Revealing Behavioural Deviations under Mutual Perception in Multi-Vehicle Autonomous Driving},
author = {Lejin Li and Xiao-Yi Zhang and Shuncheng Tang and Zhenya Zhang and Jianjun Zhao},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3744916.3787808}
}